You Can Pass This Exam For Free
Choose Your Study Path
You hold CISA/CIA/CPA and have strong audit fundamentals, but limited exposure to AI/ML concepts. You need to build AI technical knowledge before learning how to audit it.
Week 1Learn AI fundamentals: types of AI (machine learning, deep learning, NLP, LLMs, generative AI), model training approaches (supervised, unsupervised, reinforcement learning), and neural network basics
Week 2Study the AI/ML lifecycle end-to-end: data collection, feature engineering, model development, training/validation splits, deployment, monitoring, and decommissioning. Map each stage to audit control points
Week 3Deep dive into AI governance frameworks: NIST AI RMF 1.0, ISO/IEC 42001, EU AI Act risk classifications, and how they map to traditional audit frameworks you already know
Week 4Study Domain 2 Part 1 (46% of exam): data management for AI, MLOps pipelines, feature engineering, train/test splits, cross-validation, and model performance metrics
Week 5Study Domain 2 Part 2: model drift detection, bias monitoring, hallucination rates, prompt injection defenses, data poisoning, vendor/API dependencies, and SLA requirements
Week 6Study Domain 2 Part 3: change management for AI model updates, rollback procedures, incident response specific to AI failures, and supervision of AI outputs and decisions
Week 7Cover Domain 3 (21%): AI audit planning, scoping AI systems, control testing methodologies, evidence collection standards, data analytics in audits, and AI audit reporting
Week 8Practice questions across all domains. Focus on scenario-based questions and review explanations carefully. Target 70%+ on mock exams
Week 9Take full-length mock exams under timed conditions (1.6 minutes per question). Review weak areas and re-study any domains below 65%
Week 10Final review: focus on Domain 2 operations concepts, framework comparisons (NIST vs ISO vs EU AI Act), and commonly confused audit terminology
Exam Overview
Format
90 scenario-based multiple-choice questions, 150 minutes (2.5 hours). Computer-based testing at PSI testing centers or remote proctoring.
Scoring
Scaled score 200-800. Passing: 450. No penalty for wrong answers — always answer every question even if unsure.
Domains & Weights
- AI Governance and Risk33%
- AI Operations46%
- AI Auditing Tools and Techniques21%
Registration
$599 USD. Available at PSI testing centers or via remote proctoring (remote not available in India, Mainland China, or Hong Kong). Exam fee is $599 USD (non-member) or $459 USD (ISACA member). Candidates have a 6-month eligibility window after registration. Qualifying prerequisites: CISA (all holders), or CIA, US CPA, Canadian CPA, Australian CPA/FCPA, Japanese CPA, ACCA, FCCA, ICAEW ACA/FCA, CA ANZ, or Hong Kong CPA/FCPA (with IT audit or advisory role focus).
Topic Priority Table
Not all topics are tested equally. Focus your study time on Tier 1 first, then Tier 2. Tier 3 topics rarely appear — just recognize what they do.
Domain 133% of exam
AI Governance and Risk
This domain tests your ability to advise stakeholders on AI governance structures, ethical AI policies, risk assessment methodologies, data governance programs, and regulatory compliance. You must understand how AI governance maps to organizational strategy and how to evaluate whether governance controls are adequate and operating effectively.
Key Topics
NIST AI RMF 1.0ISO/IEC 42001EU AI ActAI Ethics FrameworksData GovernanceAI Risk AssessmentRACI for AI Roles
Must-Know Concepts
- AI governance structures: AI centers of excellence, AI ethics boards, RACI definitions for model owners, data scientists, risk management, and audit functions
- AI risk assessment methodologies: identify AI-specific risks (bias, drift, hallucination, adversarial attacks, data quality), assess likelihood and impact, and map to controls
- Ethical and responsible AI principles: fairness, accountability, transparency, explainability, privacy, safety, and inclusiveness — know how to audit compliance with each
- Data governance specific to AI: data lineage, data provenance, data quality metrics, data retention policies, consent management, and de-identification techniques
- Privacy considerations: data minimization, purpose limitation, consent management, cross-border data transfer restrictions, and privacy impact assessments for AI systems
- Regulatory frameworks: NIST AI RMF four functions (Govern, Map, Measure, Manage), ISO/IEC 42001 certification requirements, EU AI Act risk tiers and conformity assessments
- AI policy components: acceptable use policies, sanctioned vs unsanctioned AI tools, model approval workflows, data classification requirements, and third-party AI usage policies
- Workforce impact assessment: evaluating how AI implementation affects job roles, required skills, and organizational change readiness
- Third-party and vendor governance: right-to-audit clauses, vendor AI risk assessments, SLA requirements, and due diligence for AI service providers
Common Exam Traps
NIST AI RMF is VOLUNTARY guidance. ISO/IEC 42001 is a CERTIFIABLE standard. EU AI Act is LEGALLY ENFORCEABLE. Know which is which — the exam tests this distinction repeatedly
The EU AI Act BANS unacceptable-risk AI entirely (social scoring, real-time biometric surveillance). High-risk AI is NOT banned but requires strict conformity assessments
Data lineage (transformation history) and data provenance (origin and ownership) are DIFFERENT concepts. The exam tests whether you can distinguish them in audit scenarios
AI governance is not just IT governance applied to AI. AI introduces unique risks (bias, hallucination, drift) that require AI-specific governance controls
An AI ethics board ADVISES on ethical considerations. An AI center of excellence IMPLEMENTS standards and best practices. They serve different governance functions
Domain 246% of exam
AI Operations
The largest domain at 46% — expect roughly 41 questions on operational AI topics. Covers the entire AI solution lifecycle, data management, MLOps, testing techniques, drift and bias monitoring, AI-specific threats and vulnerabilities, change management, and incident response. This is where traditional IT auditors lose the most points, as it requires understanding AI-specific technical operations that differ significantly from conventional IT systems.
Key Topics
MLOpsFeature EngineeringModel ValidationDrift DetectionBias MonitoringAdversarial TestingAI Incident ResponseChange Management
Must-Know Concepts
- AI/ML lifecycle stages from an audit perspective: business case, data collection, data preparation, feature engineering, model development, training, validation, deployment, monitoring, maintenance, and decommissioning — know audit control points at each stage
- Data management for AI: data ingestion pipelines, feature engineering and feature stores, train/test/validation splits, cross-validation techniques, and data quality assurance
- Model development methodologies: supervised, unsupervised, and reinforcement learning approaches, model selection criteria, hyperparameter tuning, and reproducibility requirements
- Testing techniques: unit testing for AI components, integration testing, fairness testing (demographic parity, equal opportunity, disparate impact analysis), adversarial testing, and performance benchmarking
- Deployment and monitoring: model serving infrastructure, A/B testing, canary deployments, performance monitoring dashboards, drift detection (data drift, concept drift, model drift), and bias rechecking schedules
- AI-specific threats: data poisoning, prompt injection, model inversion, model theft/extraction, adversarial examples, supply chain attacks on pre-trained models, and hallucination risks
- Change management for AI: model versioning, approval workflows for model updates, impact assessment for retraining, rollback procedures, and documentation requirements for model changes
- Incident response for AI: escalation procedures for model failures, automated rollback triggers, root cause analysis for AI-specific incidents (drift, bias emergence, adversarial compromise), and communication protocols
- Vendor and API dependency management: third-party model risks, SLA monitoring, API versioning risks, and vendor lock-in assessment
- Supervision of AI outputs: human-in-the-loop requirements, confidence score thresholds, output validation procedures, and escalation criteria for low-confidence decisions
Common Exam Traps
Data drift and concept drift are DIFFERENT. Data drift means input distributions changed. Concept drift means the relationship between inputs and outputs changed. Different detection methods and remediation strategies apply
Model validation (pre-deployment testing) and model monitoring (post-deployment tracking) are DIFFERENT phases with different controls. The exam tests whether auditors verify BOTH exist
Cross-validation is a model evaluation technique, NOT a production monitoring tool. It is used during development to estimate model performance, not after deployment
Adversarial testing must be performed BEFORE deployment. It is a proactive security measure, not a reactive incident response activity
AI change management requires MORE than traditional IT change management. Model retraining can subtly change behavior without code changes, so traditional code-review-only change processes are insufficient
Hallucination monitoring is an AI-SPECIFIC control that has no equivalent in traditional IT auditing. Auditors must verify that hallucination rates are measured and acceptable thresholds are defined
Domain 321% of exam
AI Auditing Tools and Techniques
This domain covers the practical methodology of conducting AI audits, including scoping, planning, testing approaches, evidence collection, data analytics, and reporting. It also covers how AI tools can be used within the audit process itself to improve efficiency while maintaining independence and objectivity.
Key Topics
Audit PlanningControl TestingEvidence CollectionData AnalyticsAudit ReportingAI-Assisted AuditingSampling Methodologies
Must-Know Concepts
- AI audit planning and design: defining audit objectives, identifying AI system boundaries, selecting appropriate control frameworks, stakeholder identification, and risk-based scope determination
- Control testing methodologies for AI: walkthroughs, configuration reviews, code reviews, output sampling, reperformance of model predictions, and fairness/bias testing
- Evidence collection standards: sufficiency (enough evidence), reliability (trustworthy sources), relevance (related to control objectives), and reproducibility (can be independently verified)
- AI-specific audit evidence: model training logs, version control histories, performance metric dashboards, drift detection alerts, fairness test results, data lineage documentation, and approval records
- Audit sampling for AI: risk-based sampling vs random sampling, full-population testing enabled by AI analytics, stratified sampling across model outputs, and sample size determination
- Data analytics in AI audits: using AI tools for anomaly detection, pattern recognition, natural language processing for document review, and automated compliance checking
- Independence and objectivity when using AI tools in audits: avoiding over-reliance on AI-generated audit findings, maintaining professional skepticism, and documenting AI tool limitations
- AI audit reporting: communicating findings to technical and non-technical stakeholders, risk rating AI-specific findings, recommending remediation for AI control deficiencies, and follow-up procedures
- Documentation standards: workpaper requirements for AI audits, evidence preservation, audit trail completeness, and quality assurance review
Common Exam Traps
Using AI tools in the audit process does NOT mean you are auditing AI systems. These are opposite directions: Domain 2 is about auditing AI, Domain 3 is about using AI as an audit tool
Full-population testing with AI analytics does NOT eliminate the need for professional judgment. AI can flag anomalies, but the auditor must still evaluate whether findings represent actual control deficiencies
AI-assisted audit analytics must preserve auditor INDEPENDENCE. If the same AI system being audited provides the analytics tools, there is a conflict of interest
Audit evidence from AI systems must be REPRODUCIBLE. If model outputs cannot be reproduced due to stochastic behavior, the auditor must document this limitation
Risk-based sampling is generally preferred over random sampling for AI audits because AI systems have non-uniform risk distributions — certain model behaviors carry higher risk
Concepts You Must Not Confuse on the AAIA Exam
These pairs appear on nearly every exam. Learn the difference and you'll avoid the most common traps.
Top Mistakes to Avoid
Confusing NIST AI RMF (voluntary guidance) with ISO/IEC 42001 (certifiable standard) with EU AI Act (enforceable law) — each has a fundamentally different compliance status
Mixing up data drift (changing input distributions) with concept drift (changing input-output relationships) — they require different detection and remediation approaches
Treating model validation (pre-deployment testing) and model monitoring (post-deployment tracking) as interchangeable — both must exist independently
Confusing data poisoning (attacks training data before deployment) with prompt injection (attacks model inputs after deployment) — different attack phases require different controls
Assuming traditional IT change management is sufficient for AI systems — model retraining can change behavior without code changes, requiring AI-specific change controls
Mixing up data lineage (how data was transformed) with data provenance (where data originated) — the exam tests both in governance and audit scenarios
Thinking that using AI in the audit process is the same as auditing AI systems — Domain 2 audits AI systems while Domain 3 uses AI as an audit tool
Confusing explainability (technical: how the model decides) with transparency (organizational: disclosing AI usage) — auditors must verify both separately
Assuming that passing model validation once means the model is permanently safe — models degrade over time due to drift, requiring continuous monitoring
Forgetting that AI audit evidence must be reproducible — stochastic model behavior can produce different outputs for the same input, which must be documented
Exam-Ready Checklist
Can explain all 3 exam domains and their relative weights (33%, 46%, 21%) and allocate study time accordingly
Understand the complete AI/ML lifecycle and can identify audit control points at every stage from data collection through decommissioning
Can distinguish between NIST AI RMF (voluntary), ISO/IEC 42001 (certifiable), and EU AI Act (legally binding) and know when each applies
Know all three types of drift (data, concept, model) and can identify appropriate detection and remediation controls for each
Can apply fairness testing methodologies including demographic parity, equal opportunity, and disparate impact analysis in audit scenarios
Understand AI-specific threats (data poisoning, prompt injection, model inversion, adversarial examples) and can verify controls exist for each
Know the difference between auditing AI systems and using AI tools in the audit process, and can maintain independence when using AI-assisted analytics
Can evaluate AI change management controls including model versioning, approval workflows, rollback procedures, and retraining documentation
Understand AI audit evidence standards: sufficiency, reliability, relevance, and reproducibility for AI-specific artifacts like training logs and fairness reports
Can assess vendor/third-party AI risks including right-to-audit clauses, SLA compliance, and data sharing governance
Know AI incident response procedures and how they differ from traditional IT incident response (drift escalation, bias emergence, model failure rollback)
Scored 65%+ on at least two full mock exams under timed conditions (1.6 minutes per question, 450/800 passing score)
Recommended Resources
Free & Official Resources
Paid Courses & Practice Exams
These are recommended if you prefer a structured learning path. They can save time but are not required to pass.